SMART2: Multi-Library Statistical Mitogenome Assembly with Repeats Fahad Alqahtani1;2 and Ion I. Măndoiu1 1 Computer Science & Engineering Department, University of Connecticut, Storrs, CT, USA, {fahad.alqahtani,ion.mandoiu}@uconn.edu 2 National Center for Artificial Intelligence and Big Data Technology, King Abdulaziz City for Science and Technology, Riyadh, Saudi Arabia Abstract. SMART2 is an enhanced version of the SMART pipeline for mitogenome assembly from low-coverage whole-genome sequencing (WGS) data. Novel features include automatic selection of the opti- mal number of read pairs used for assembly and the ability to assem- ble multiple sequencing libraries when available. SMART2 succeeded in generating mitochondrial sequences for 26 metazoan species with WGS data but no previously published mitogenomes in NCBI databases. The SMART2 pipeline is publicly available via a user-friendly Galaxy inter- face at https://neo.engr.uconn.edu/?tool_id=SMART2. Keywords: Mitogenome assembly, multi-library assembly, low-coverage sequenc- ing 1 Introduction Mitochondria are cellular organelles present with very rare exceptions in all eukaryotic cells. In most animals, the mitochondria have their own genome, a double-stranded circular DNA molecule typically ranging in size between 15-20Kb that encodes 37 genes (2 ribosomal RNA genes, 13 protein coding genes, and 22 transfer RNA genes). The mitochondrial genome is inherited maternally, and has much higher copy number than the nuclear genome [24]. The small size, high copy number, and the presence of both coding and regulatory regions that mutate at different rates make the mitochondrial genome an ideal genetic marker. Indeed, mitochondrial sequences have been used in applications ranging from maternal ancestry inference and tracing human migrations [6] to forensic analysis [19]. The mitochondrial DNA has also become the workhorse of biodiversity studies since many non-model species do not have yet a sequenced nuclear genome [12,16]. To date, most such biodiversity studies have been based on sequencing a single gene fragment, such as the Cytochrome C oxidase I (COI) gene, which has been adopted as the preferred “barcode of life” [14,21]. Recently there have been a renewed appre- ciation for the improved accuracy of taxonomic and phylogenetic analyses performed based on complete mitogenome sequences assembled from low coverage whole genome shotgun (WGS) reads generated using next generation sequencing (NGS) technologies. Indeed, full length mitogenome sequences capture evolutionary events such as genome rearrangements that are missed in single gene analyses [18]. Furthermore, the expo- nential decrease in NGS costs has led to an explosion in the number of WGS datasets generated from non-model organisms. For mammals alone, there are currently over two hundred species with paired-end WGS data available in the NCBI SRA database but for which no complete mitogenome is available. Recent studies have also demonstrated that WGS data of sufficient depth for reconstructing mitogenomes can be generated from preserved museum specimens [23], making the approach applicable to rare or even extinct species. Leveraging the available WGS datasets to expand the number of complete mi- togenomes requires bioinformatics pipelines that can assemble and annotate high- quality mitogenomes quickly and with minimal human intervention. Unfortunately, standard genome assemblers often fail to generate high quality mitochondrial genome sequences due to the large difference in copy number between the mitochondrial and nuclear genomes [13]. This has led to the development of specialized tools for recon- structing mitochondrial genomes from WGS data, mainly falling within three cate- gories. Reference-based methods such as MToolBox [8] require the mtDNA sequence of the species of interest or a closely related species, which are often not available for the less-studied species of interest in biodiversity studies. Seed-and-extend tools such as MITObim [13] and NOVOPlasty [11] use a greedy approach to extend available seed sequences such as the COI but can have difficulty handling repetitive regions present in some mitochondrial genomes [16]. Finally, de novo methods such as Norgal [2] and plasmidSPAdes [7] use coverage-based filtering to remove nuclear WGS reads before performing assembly using the de Bruijn graph of remaining reads. In [5] we introduced a hybrid method called Statistical Mitogenome Assembly with RepeaTs (SMART), which uses a seed sequence to estimate the mean and standard deviation of mtDNA k-mer counts, then positively selects reads with k-mer counts falling within three standard deviations of the estimated mean before performing de novo assembly. Experiments in [3] show that for low-depth WGS datasets the positive selection approach implemented by SMART yields higher enrichment for mtDNA reads than the negative selection of Norgal. Furthermore, SMART was shown to produce complete circular mitogenomes with a higher success rate than both seed-and-extend tools MITObim and NOVOPlasty and de novo assemblers Norgal and plasmidSPAdes. In this paper we present an extension of the SMART pipeline, referred to as SMART2, that can take advantage of multiple sequencing libraries when available and automatically selects the optimal number of read pairs used for assembly. We also present experimental results comparing read filtering and assembly accuracy of SMART2 with that of existing state-of-the-art tools, along with the results of a pilot “orphan mitogenomes” project in which SMART2 was used to generate 15 complete and 11 partial mitogenomes for 26 mammals and amphibians without previously pub- lished mitogenomes. All novel mitogenomes have been submitted to GenBank as Third Party Annotation (TPA) sequences [9]. 2 Methods The SMART2 pipeline is deployed using a customized instance of the Galaxy framework [1] and is publicly available via a user-friendly Galaxy interface at https://neo.engr. uconn.edu/?tool_id=SMART2 (see Fig. 1). The pipeline was designed for processing paired-end reads in fastq format from one or two WGS libraries. In addition to fastq files, the user specifies the sample name and a seed sequence in fasta format. By default Fig. 1. Galaxy interface of SMART2. the number of reads is selected automatically as described below, but the user can override the default and manually specify it. Advanced options also allow the user to change the default choices for the number of bootstrap samples (default is 1), k-mer size (default is 31), number of threads (default is 16), and the genetic code used for MITOS annotation (default is the vertebrate mitochondrial code). The main steps of the SMART2 pipeline follow those of SMART with adaptations for multi-library inputs: 1. Automatic adapter detection and trimming, performed independently for each li- brary. 2. Random resampling of a number of trimmed read pairs, either specified by the user or automatically determined using the doubling strategy described below. 3. Selection of mitochondrial reads based on coverage estimates of seed sequence k- mers – aggregated across libraries using one of the methods described below (2- dimensional Gaussian mixture modeling using MCLUST, Union, or Intersection). 4. Joint preliminary assembly of reads passing the coverage filter in the two libraries, performed using SPAdes. 5. Filtering of preliminary contigs by BLAST searches against a local mitochondrial database. 6. Secondary read filtering by alignment to preliminary contigs that have significant BLAST matches, performed independently for each library. PE WGS Library 1 Seed Sequence PE WGS Library 2 Trimmed Reads 1 Adapter Detection Adapter Detection Trimmed Reads 2 & Trimming & Trimming Fit 2-Component Random Bootstrap Reads Bootstrap Reads Random GMM to Seed Resampling Sample 1 Sample 2 Resampling K-mer Counts Sufficient No: double size No: double size Coverage? Yes Coverage-Selected Coverage-Based Coverage-Selected Reads Read Filter Reads SPAdes Assembler Preliminary BLAST Filtered Contigs Contigs HISAT2 HISAT2 Build HISAT2 Align Reads Contigs Index Align Reads Alignment-Selected SPAdes Alignment-Selected Reads Assembler Reads Assembly No Graph No Eulerian? Yes Maximum Likelihood Path Search Repeat N times Pairwise Fitting Cluster Annotated Scaffold MITOS Annotation Alignment & Consensus Consensus Sequences Pipeline Clustering Sequences Sequences Fig. 2. SMART2 workflow. Coverage-based reads Filter: Mean Standard deviation 26.96 6.91 COI kmers counts distribution Gaussian Mixture Model for COI K-mers Counts of Two samples Figure - Gaussian Mixture Model for COI K-mers Counts of Two samples. Figure - COI kmers counts distribution. (a)Filtered mtDNA reads: (b) Filtered mtDNA reads: Filtered mtDNA #Read pairs Read average length (bp) Data Reads pairs Read average length (bp) Fig. 3. Mitochondrial k-mer coverage distributionSample#1 estimated 7,513 by MCLUST 100.72 using seed Filtered mtDNA 6,780 100.84 k-mer counts generated from (a) 800k readSample#2 pairs sampled from 7,275 library SRR630623 100.69 of Preliminary assembly: the Anopheles stephensi dataset, and (b) 400kPreliminary read assembly: pairs sampled from each of the two Number of contigs longest linear contig libraries of the Anopheles stephensi dataset.Number of contigs longest linear contig 197 4,937 291 8,628 Preliminary contig filtering: Preliminary contig filtering: Number of contigs longest linear contig Number7. of Jointcontigs secondary assembly longest linear of contig selected reads, performed using SPAdes.